Ccg Analyze

ccg-analyze — Code Analysis

Graph-based analysis for change impact, call flow, dead code, module structure — anything about "relationships."

Intent → Tool Mapping

| User intent | Tool | Notes | | ------------------------------------ | ------------------------------------------------ | --------------------------------------------- | | "Impact of changing this function?" | get_impact_radius (start with depth 3) | Widen to depth 5 if too narrow | | "Trace call flow from this function" | trace_flow | If broken at interfaces, see workaround below | | "Who calls this function?" | query_graph (callersof) | | | "What does this function call?" | query_graph (calleesof) | | | "Unused code" | find_dead_code | Scope by path/kind on large namespaces | | "Large functions" | find_large_functions | Use limit; DB still scans matches first | | "Fallback edge quality" | find_suspect_fallback_edges | Page and inspect as a quality report | | "Risk of this change" | detect_changes + get_affected_flows | git diff-based | | "Module structure" | list_communities + get_architecture_overview | Use pagination on large namespaces | | "Test coverage gaps" | get_community (with coverage) | Page members if include_members=true |

trace_flow Limitations & Workaround

If trace_flow returns only 1–2 starting nodes, it broke at an interface dispatch boundary (Go's h.deps.X.Method(), Python duck typing, etc.).

Workaround:

1. Confirm starting point via trace_flow
2. Manually expand graph with callers_of / callees_of
3. When you hit interface methods:
   - Find the interface via search
   - Reverse-trace implementations with callers_of

Latest ccg added an interface dispatch resolver, but it doesn't cover every pattern. If results look thin, fall back to this manual pattern.

getimpactradius Tips

• depth 1–2: direct impact (immediate callers/callees)
• depth 3: recommended default — covers most tasks
• depth 5+: large monorepo propagation. Watch for noise.

If results are huge, the change scope is likely too wide. Reconsider the change unit based on node count.

Pagination Defaults

Use explicit budgets for graph browsing tools when the namespace may be large:

| Tool | Parameters | Default starting point | | ---- | ---------- | ---------------------- | | query_graph | limit, offset | limit=50, offset=0 | | list_flows | limit, offset | limit=50, offset=0 | | list_communities | limit, offset | limit=50, offset=0 | | find_suspect_fallback_edges | limit, offset | limit=50, offset=0 | | get_community | member_limit, member_offset | Use only when include_members=true | | get_architecture_overview | community_limit, community_offset, coupling_limit, coupling_offset | Start with 50 each |

Paginated responses include has_more. If true, call again with next_offset. Do not request the max page size first for LLM analysis; use 50 or 100 unless the user specifically needs a bulk export.

High-Volume Tool Caution

These tools are useful, but can produce large responses in real services:

• find_dead_code: scope with path and kinds before broad scans.
• find_suspect_fallback_edges: inspect as an operational quality report, not a default first step.
• find_large_functions: pass limit, and prefer path when reviewing one module.
• Architecture/onboarding prompts: use after list_communities or get_architecture_overview has confirmed the namespace size.

Accuracy Limits (use with awareness)

• Interface calls may over-predict (expands to all implementations)
• Dynamic dispatch (reflection, plugins) → not captured
• Build-tag-split files → both registered (noise)
• Fallback call edges improve recall but may add false positives; use strict mode when evidence quality matters more than coverage
• Don't trust 100%. For important decisions, cross-check with grep.

Full MCP Tool List

| Tool | One-liner | | --------------------------- | ---------------------------- | | get_impact_radius | BFS blast radius | | trace_flow | Call chain trace | | find_large_functions | Above line threshold | | find_dead_code | No callers | | find_suspect_fallback_edges | Fallback call-edge quality report | | detect_changes | Git diff risk score | | get_affected_flows | Flows affected by change | | list_flows | Stored flow list, paginated | | list_communities | Louvain module clusters, paginated | | get_community | Community details + coverage, paginated members | | get_architecture_overview | Coupling summary, paginated |

For detailed parameters, see MCP schema.

Prerequisites

Requires ccg build . first. Schema error → ccg migrate, retry. (See /ccg skill.)

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: tae2089
• Source: tae2089/code-context-graph
• License: MIT

Install and usage instructions live in the source repository linked above.